A Model Officer: An Agent-based Model of Policing
Sarah Wise*1 and Tao Cheng1
Department of Civil, Environmental, and Geomatic Engineering
University College London
November 7, 2014
Summary
The way police officers create guardianship is poorly understood, in part because of the
complexities of policing. However, in order to understand how to advise the police,
researchers must have an understanding of how the current system works. The work presents
an agent-based model that simulates the movement of police vehicles, using a record of calls
for service to emulate the demands on the police force. The GPS traces of the simulated
officers are compared with real officer movement GPS data in order to assess the quality of
the generated movement patterns.
KEYWORDS: agent-based modelling, crime, policing, GIS
1. Introduction
The term guardianship is a criminological concept that refers to the way guardians, such as
property owners and the police, prevent potential offenders from committing crimes (Cohen
and Felson, 1979). When potential offenders are choosing whether to commit a crime, they
consider how likely they are to be apprehended or stopped by any fellow citizens or police
officers in their immediate area (Kleck and Barnes, 2008). The offender’s choice to offend is
therefore based in part on his or her interactions with other people, and the higher-level crime
patterns that result from the choices of all of the individual offenders are influenced by the
physical presence (or absence) of police and citizens. Thus, guardianship depends on the
spatial, temporal, and behavioural interactions of offenders, citizens, and police officers.
However, the way police create guardianship is not obvious. Policing is a complex, culturally
specific process, and officers have many intersecting responsibilities (Policy Studies
Institute, 1996). While Robert Peel identified the prevention of crime and disorder as the first
goal of policing (Home Office, 2012), police forces are also asked to help with finding
missing persons, providing security at public festivals, and handling traffic accidents
(Metropolitan Police, 2014). For researchers attempting to influence crime rates by
suggesting police policy, ignorance of the way these other commitments constrain officer
presence and movement will result in policy suggestions divorced from reality.
Researchers have historically failed to consider these complications in their models of
policing and guardianship. Throughout the literature, very few models of guardianship
creation exist, and those that do explore it in trivial ways. In general, the issue these models
have faced is two-fold: firstly, that many methodologies cannot aggregate lower-level
behaviours in order to understand the purposive behaviour of a system comprised of many
individuals, and secondly that there is an absence of data that could support such a model. A
1 s.wise@ucl.ac.uk number of simulations emulate the process of officers carrying out responsibilities as part of
a larger group, but ignore the many complicating factors, such as the purposeful movement
of officers and the fact that they have other responsibilities (e.g. Birks et al., 2012; Groff,
2007). Further, simulated officers are unimpeded by the time-consuming process of actually
dealing with offenders (e.g. Melo et al., 2006; Dray et al., 2008) These simplifications
significantly bias officer movement patterns, generating patterns of guardianship that do not
match the guardianship created by real officers.
In all of these cases, researchers have been hampered by a lack of access to information
about officer duties, incapable of incorporating the complexities of policing into their
simulations for want of data. As a result of our working relationship with the London
Metropolitan Police, we have access to this kind of information. This work presents a
simulation which seeks to capture the complex realities of policing, using a combination of
data and behavioural research to create a realistic model of police activity. Given that
policing is complex, spatial, temporal, and profoundly influenced by human decisionmaking, we utilise an agent-based model (ABM).
2. The Model
The model presented here utilises an agent-based framework to explore how officers
translate their assignments into movement and actions in the context of the environment in
which they find themselves. As a methodology, ABM has been particularly successful in
incorporating criminological concepts such as routine activity theory (Cohen and Felson,
1979), rational choice theory (Cornish and Clarke, 1987), and crime pattern theory
(Brantingham and Brantingham, 1984) into simulations (e.g. Groff, 2006; Birks et al., 2012;
Malleson et al., 2012).
In this work, the ABM attempts to capture the behaviours of Metropolitan Police constables,
simulating them at the level of the vehicles to which they are assigned. The simulation
models the vehicles moving over a road network, specified with 1m2 resolution, and are
updated on a temporal scale of one minute per simulation step. The model framework is built
in Java, using the MASON simulation toolkit, an open-source multiagent simulation library.
The following sections will describe the environment in which the vehicles exist, the way
vehicles are represented in the simulation, and the way vehicle behaviours are translated into
actions
2.1 Environment
In order to represent the environment in which police vehicles exist, the model combines
information about the real-world road network with records of calls for service from the
community. The model was tested with road network data derived from the Ordnance Survey
MasterMap Integrated Transport Network Road (ITN) dataset. The locations of police
stations, which factor into the activities of the police vehicles, are taken from the data
provided to us by the Met Police. The locations of traffic lights, which impose a time cost on
the movements of officers throughout the environment, is taken from data provided by
Transport for London (TfL).
In addition to the physical constraints of the environment, vehicles are influenced in how
they move by the calls for service that they receive from the general public. The timing and
location of these calls for service are drawn from the records of the Call Aided Despatch
(CAD) system of the Met Police, and these incidents are used to direct officers to the real-
world locations of incidents at the appropriate times. Thus, the pressures and constraints
upon the officers are rendered in a simulated setting.
2.2 Agents
The model represents the actions of police officers in terms of the movement and interactions
of police vehicles. Vehicle agents have attributes that inform their actions. Table 1 provides
an overview of the attributes that characterise the Vehicle agent at any given point in time,
specifying the range of values these attributes may take on and providing examples of such
values. In particular, Vehicles have a current location in space, a home station, a unique call
sign, a current activity, and a current status, as well as a Tasking object.
Table 1. Vehicle attributes
All officers obey a daily schedule of returning to their home station every eight hours, to
simulate the change in officers. In addition to this shared structure, individual vehicles are
assigned to “taskings”, or assignments of duty. The vehicle’s assignment corresponds to the
real-world process of assigning officers to carry out different tasks during the day, as is
arranged during the briefing before a shift begins. These assignments dictate who will
respond to calls, who will focus on patrolling, who will be responsible for coordinating with
other officers in order to pick up offenders, and so forth. They structure the officer’s day, and
dictate how he progresses from one activity to another. The existing assignments are:
•
Reporting: the vehicle is primarily responsible for responding to non-urgent calls
for service. It will move around the environment in response to these calls, spending
time dealing with the caller when it reaches the site of the incident. The vehicle will
spend any unoccupied time patrolling, which is modelled here as moving randomly
about the environment.
•
Transporting: the vehicle is responsible for coordinating with other vehicles who
have detained suspects. When the Transporting vehicle receives a request of
transport, it will move to the suspect and then transport him back to the police
station.
•
Responding: the vehicle is responsible for responding to urgent calls for service.
When it receives a call, the responding vehicle begins to move at a faster speed and
ignores streetlights, both in its calculation of the shortest path and in its movements.
When it reaches the site of the incident, it spends time assessing the incident,
potentially apprehending an offender. If an offender is apprehended, the vehicle will
call for a Transport vehicle and wait until it arrives. It will spend any unoccupied
time patrolling.
Figure 1. Normalised heatmaps showing the road usage associated with the real data (A), the
random-patrolling model (B), and the tasking-service model (C).
The vehicles communicate with an object called the Despatcher, which informs them of
incidents based on the data fed into the simulation and coordinates among the agents, taking
a request for transport from a Responding vehicle and transmitting the request to a Transport
vehicle. Vehicles plan the shortest path to a point in terms of time, and obey traffic lights
except in the case of responding vehicles moving to an urgent request.
3. Results
In order to investigate how the existence of assignments influences the behaviour of
simulated agents, we compare two models: one in which all the vehicles spend their time
patrolling and receive no calls for service, and another in which vehicles are assigned
taskings and receive calls for service as drawn from the real CAD data. The first model
emulates the existing ABMs of police movement, while the second represents our
contribution.
Figure 1 shows the real road usage data compared with the random-patrolling model and the
tasking-service model. Briefly, the real data shows officers making greater use of major
roads. The random-patrolling model demonstrates a much less concentrated focus on these
roads, while the tasking-service model is more concentrated. However, the tasking-service
model is more focused in the south and centre of the region than is the real data. The
simulated models generate many more records than exist in the real data, which poses
interesting questions about how to compare synthetic versus real data.
4. Conclusions
The simulation generates interesting results, with the behavioural model generating more
realistic patterns of road usage than the commonly utilised random movement model. More
questions exist with regard to comparing the real data with the generated data, and suggest
further investigations into the growing field of ABM validation efforts. The work presented
here both addresses the lack of nuanced simulations of policing and pushes forward the
practice of inserting real-world data into simulations in order to emulate rich environments
for behaviourally complex agents. In the future, this ABM will allow us to explore
counterfactual situations, comparing the projected effectiveness of different policing
strategies.
5. Biography
Sarah Wise is a postdoctoral researcher in the Department of Civil, Environmetnal, and
Geomatic Engineering at University College London. Her research interests include agentbased modelling, social network analysis, data mining, and geographical information
systems, and her past research has dealt with crisis situations, health, crime, and social
media.
Tao Cheng is a Professor in GeoInformatics, and Director of SpceTimeLab for Big Data
Analytics (http://www.ucl.ac.uk/spacetimelab), at University College London. Her research
interests span network complexity, Geocomputation, integrated spatio-temporal analytics and
big data mining (modelling, prediction, clustering, visualisation and simulation), with
applications in transport, crime, health, social media, and environmental monitoring.
References
Birks, D., Townsley, M., & Stewart, A. (2012). Generative Explanations of Crime: Using
Simulation To Test Criminological Theory. Criminology, 50(1), 221–254.
Brantingham PJ and Brantingham PL, 1984, Patterns in crime. Macmillan, New York, NY.
Cornish D and Clarke R, 1987, Understanding crime displacement: An application of
rational choice theory. Criminology, 25(4): 933–947.
Cohen L and Felson M, 1979, Social Change and Crime Rate Trends: A Routine Activity
Approach. American Sociological Review, 44(4): 588–608.
Dray, A., Mazerolle, L., Perez, P., & Ritter, A. (2008). Policing Australia’s “heroin
drought”: using an agent-based model to simulate alternative outcomes. Journal of
Experimental Criminology, 4(3), 267–287.
Groff, E. R. (2007). “Situating” Simulation to Model Human Spatio-Temporal Interactions:
An Example Using Crime Events. Transactions in GIS, 11(4), 507–530.
Home Officer. (2012). Policing by Consent.
https://www.gov.uk/government/publications/policing-by-consent
Kleck, G., & Barnes, J. C. (2008). Deterrence and Macro-Level Perceptions of Punishment
Risks: Is There a “Collective Wisdom”? Crime & Delinquency, 59(7), 1006–1035.
Melo, A., Belchior, M., & Furtado, V. (2006). Analyzing Police Patrol Routes by Simulating
the Physical Reorganization of Agents. In J. S. Sichman & L. Antunes (Eds.), MultiAgent-Based Simulation VI (pp. 99–114). Springer Berlin Heidelberg.
Metropolitan Police. (2014) Specialist Crimes and Operations,
http://content.met.police.uk/Site/specialistcrimeoperations